vet402

Independent verification of the x402 agent-payment economy.

We buy. We settle. We publish the measurements.

KIZUNA Creation — Measure Reality. Build Trust.

[vet402 @vet_402]

Vouch is now vet402.

Same builder, sharper mission: independent verification of the x402 agent-payment economy.

We buy what endpoints actually sell. We settle on-chain. We publish the measurements — with tx hashes, not estimates.

[vet402 @vet_402]

We could have shipped the contrast fixes and moved on. Instead the gate now fails the build if a banned rule reappears in shipped CSS — because it did reappear, five times, resurrected by our own deploy. Decisions you don't enforce aren't decisions.

https://t.co/HA1buqXT89

[vet402 @vet_402]

Why “402”: HTTP 402 Payment Required waited 28 years for a real job. x402 gave it one.

vet402 checks the work.

[vet402 @vet_402]

Our budget guard is fail-closed: a hard daily cap enforced in code. Spending halts when the ledger says stop.

A verification institution that can't govern its own spending has no business measuring anyone else's.

[vet402 @vet_402]

Hi @murrlincoln — following up on the micro-grants note in coinbase/x402 PROJECT-IDEAS.md.

vet402 makes real purchases against listed x402 endpoints daily on Base mainnet (745 paid attempts, every receipt with tx hash) and publishes the data free.

Demo video: https://t.co/h5p70pecES

Happy to share more if useful!

[vet402 @vet_402]

We buy from x402 endpoints ourselves and publish what settled — with the on-chain tx hash. 17,572 endpoints observed daily, 745 real purchases on Base mainnet. Every receipt is machine-readable. Live at https://t.co/L1WHr5fOSv @coinbaseDev

[vet402 @vet_402]

Cold-probing a POST endpoint with a GET marks a healthy business as dead (x402 repo #3113).

Our prober respects the declared method. Small detail — entire revenues have silently disappeared over it.

[vet402 @vet_402]

We buy from x402 endpoints ourselves and publish what settled — with the on-chain tx hash. 17,572 endpoints observed daily, 745 real purchases on Base mainnet. Every receipt is machine-readable. Live at https://t.co/L1WHr5fOSv @coinbaseDev

[vet402 @vet_402]

ERC-8004 has three registries.

Identity: 173k rows.
Reputation: 155k entries — 98.7% with no verifiable transaction behind them.
Validation: effectively empty.

We intend to fill the empty one, with records of real purchases.

[vet402 @vet_402]

State of x402 (measured, not estimated):
17,167 endpoints on record, 90.2% still listed.
Only 4.5% answer a valid 402 when we probe them.
We paid 745 of them for real — 279 settled with an on-chain receipt.
https://t.co/txesrsAxRi

[vet402 @vet_402]

Roughly half of observed x402 traffic is self-dealing (Artemis, 2026). Payment history can be manufactured for cents.

The only signal a seller cannot fake is a purchase they didn't know was a test.

We control our own sampling.

[vet402 @vet_402]

A failing verdict publishes with raw log, tx hash, timestamp and reproduction steps — or it doesn't publish.

Sellers can dispute any result and trigger a free re-verification. Corrections are logged publicly.

That's the whole design, in three sentences.

[vet402 @vet_402]

Verification levels we build on:

L0 — does the endpoint answer correctly?
L1 — does payment settle and a response arrive?
L2 — does the response match the seller's own declaration?
L3 — opinion. Never mixed with L0–L2.

Facts and opinions get separate tables.

[vet402 @vet_402]

Every number here is public, with raw logs and reproduction steps attached — not a score.

https://t.co/1m8iMI21ej

We buy. We settle. We publish the measurements.

[vet402 @vet_402]

We ran the x402 Observatory's first full day of real purchases. 531 payments signed and sent to endpoints in the public catalog. 205 settled with an on-chain receipt. 38.6%.

[vet402 @vet_402]

Catalog-wide (L0, no purchase attached): 15,092 endpoints on record. 549 have a payment wall confirmed to answer a valid 402 on the declared method. The other 14,543 have not been reached yet by the rolling probe.

[vet402 @vet_402]

Catalog-wide (L0, no purchase attached): 15,092 endpoints on record.

549 have a payment wall confirmed to answer a valid 402 on the declared method.

The other 14,543 have not been reached yet by the rolling probe.

[vet402 @vet_402]

We ran the x402 Observatory's first full day of real purchases.

531 payments signed and sent to endpoints in the public catalog.
205 settled with an on-chain receipt.
38.6%.

[vet402 @vet_402]

The x402 catalog problem: 14,865 listings, about 3.5% with any sign of real demand (independent survey, July 2026).

An index that can't tell the difference isn't discovery.

Our observatory starts with the boring part: measuring liveness, daily, for every listing.

[vet402 @vet_402]

Why: proof of payment exists on-chain. Proof of delivery exists nowhere.

98.7% of ERC-8004 reputation feedback has no verifiable transaction behind it (arXiv:2606.26028).

Reading registries isn't verification. Buying is.

[vet402 @vet_402]

What changes: a continuous observatory of x402 endpoints, verification levels L0–L2, and every failing verdict published with raw log + tx hash + reproduction steps.

What stays: Verified Payee, the API, the public accuracy ledger.

Building in public.

[vet402 @vet_402]

Vouch is now vet402.

Same builder, sharper mission: independent verification of the x402 agent-payment economy.

We buy what endpoints actually sell. We settle on-chain. We publish the measurements — with tx hashes, not estimates.